Data Mining Innovation to Predict the Duration of Tax Disputes

A Case Study in Indonesia

Abstract

The resolution of complex tax disputes poses a significant challenge to Indonesia's tax system, impacting administrative efficiency and taxpayer certainty. This study develops prediction models for tax dispute resolution duration using machine learning techniques: Decision Tree, Random Forest, and Support Vector Machine (SVM). The analysis covers 16,223 dispute cases from 2016 to 2023, employing data mining to identify critical factors influencing resolution times. Results indicate that Random Forest and SVM models achieve high accuracy (99.7%), significantly outperforming traditional methods. The Random Forest excels in interpretability, whereas the SVM delivers stable predictions compared to the Decision Tree. These findings imply potential improvements in dispute resolution speed, resource optimization, administrative transparency, and automation, thereby reducing case backlogs and enhancing taxpayer confidence. The primary contribution lies in applying machine learning to enhance Indonesia's tax dispute resolution efficiency, providing an accurate, objective, and data-driven method. This research also suggests future opportunities to develop advanced prediction models using ensemble learning or deep learning techniques. Such developments could further enhance the fairness and transparency of the tax system.